Analytical Modeling of Debonding Mechanism for Long and Short Bond Lengths in Direct Shear Tests Accounting for Residual Strength

Interfacial debonding in fiber-reinforced composites is a common problem, especially in external strengthening techniques. This investigation aims to determine the load during debonding, and discusses two practical design parameters for direct shear tests, which are commonly used to assess the mechanics of debonding. In this study, three different bond-slip cohesive laws and one finite fracture mechanics approach are considered to investigate debonding in direct shear tests by taking the effect of residual strength into account. For each model, load during debonding and its maximum value are given by closed-form expressions, which are then checked against experimental data reported in the literature. It is shown that using the interfacial mechanical properties extracted from one geometry, the debonding load of tests with different bond lengths and widths can be predicted without any fitting procedure. Moreover, effective bond length formulae are suggested for each model; one is the straightforward extension (accounting for residual strength) of a formula available in the Standards. The results illustrate the importance of considering residual strength in direct shear tests, even at debonding onset, with its effect being nonetheless higher for long bond lengths.

Machine learning-based prediction of the translaminar R-curve of composites from simple tensile test of pre-cracked samples

A novel approach to determine the translaminar crack resistance curve of composite laminates by means of a machine learning model is presented in this paper. The main objective of the proposed method is to extract hidden information of crack resistance from strength values of center-cracked laminates. Compared to traditional measurements, the notable advantage is that only tensile strength values are required which can be obtained by a rather simpler experimental procedure. This is achieved by the incorporation of the finite fracture mechanics, which links crack resistance with strength values. In order to get training dataset, a semi-analytical method using both finite element method and finite fracture mechanics is employed to generate strength values of center-cracked specimens with different random R-curves, which serve as inputs for our artificial neural network. Regarding the outputs, principal component analysis is performed to reduce dimensionality and find suitable descriptors for crack resistance curves. After successfully training machine learning model, experimental studies on basalt fiber reinforced laminates are conducted as validation. Results have proven the effectiveness of the proposed strategy for predicting crack resistance curves, as well as the feasibility of using machine learning-based framework to find out more information about composites from simple experimental data.

Non-local criteria for the borehole problem: Gradient Elasticity versus Finite Fracture Mechanics

Two nonlocal approaches are applied to the borehole geometry, herein simply modelled as a circular hole in an infinite elastic medium, subjected to remote biaxial loading and/or internal pressure. The former approach lies within the framework of Gradient Elasticity (GE). Its characteristic is nonlocal in the elastic material behaviour and local in the failure criterion, hence simply related to the stress concentration factor. The latter approach is the Finite Fracture Mechanics (FFM), a well-consolidated model within the framework of brittle fracture. Its characteristic is local in the elastic material behaviour and non-local in the fracture criterion, since crack onset occurs when two (stress and energy) conditions in front of the stress concentration point are simultaneously met. Although the two approaches have a completely different origin, they present some similarities, both involving a characteristic length. Notably, they lead to almost identical critical load predictions as far as the two internal lengths are properly related. A comparison with experimental data available in the literature is also provided.